Small disruptions cause ripple effects that destroy the economy.

WHAT are the effects of financial market disruptions on economic activity? The recent global financial crisis suggests that the effects are large and highly persistent: by 2017, ten years after the beginning of the crisis, the U.S. and U.K. GDPs remain far—10 percentage points (pp) or more—from their precrisis trends (figure 1). More systematic narrative studies of financial stress episodes point to similar conclusions. For instance, Romer and Romer (2017) study a panel of OECD countries and find that GDP is typically 9 pp lower five years after an extreme financial stress episode like the recent crisis.1 Surprisingly, however, these numbers stand in sharp contrast with the findings of another influential literature on the importance of financial markets for economic activity. Multivariate time series models (i.e., structural VARs) find relatively mild and short-lived effects of financial shocks— shocks to the effective “risk-bearing capacity” of the intermediary financial sector. For instance, the results of Gilchrist and Zakrajšek (2012) imply that output should be only 1.3 pp lower five years after an adverse financial shock like the one experienced in the recent crisis.2 To make sense of this conundrum, we first point to separate shortcomings of the two aforementioned approaches noted— narrative accounts and structural VARs. On the one hand, and unlike structural VARs, narrative accounts are not designed to identify the causal effect of financial strains on economic activity, only the existence of a correlation. On the other hand, structural VARs do not take into account that financial shocks are likely to have asymmetric effects on economic activity, as has been predicted theoretically (Mendoza, 2015; He & Krishnamurthy, 2013; Brunnermeier & Sannikov, 2014). In contrast, narrative accounts implicitly allow for asymmetric effects because they focus only on adverse financial developments: negative shocks. We then consider an empirical model, a vector movingaverage (VMA) model, designed to address these separate limitations identifying the causal effects of financial shocks and taking into account the possible asymmetric effects of financial shocks. Like VARs, they can incorporate structural identifying restrictions to tease out causal effects, but unlike VARs, they can easily be generalized to allow for asymmetric effects of shocks. Our baseline evidence is based on U.S. data, and we establish the causal effect of financial shocks by using an identification strategy that builds on but also expands (Gilchrist & Zakrajšek, 2012, GZ). We isolate innovations to the excess bond premium (EBP)—the component of credit spreads purged from the expected default risk of borrowers—that are contemporaneously orthogonal to macrovariables, and we separate the EBP innovations into monetary shocks and financial shocks using a proxy variable approach based on Romer and Romer’s (2004) narrative measure of exogenous monetary policy changes. We find that a favorable financial shock—an easing of financial conditions—has little effect on economic activity, but an adverse financial shock has large and persistent effects on economic activity. These results help reconcile the seemingly contradictory findings between narrative accounts and structural time series analyses: structural VARs have found mild and transitory effects of financial shocks on GDP because VARs are linear models, in which the large and persistent effects of adverse shocks are mixed with the (according to our results) small and transitory effects of favorable shocks, leading to mild average effects of financial shocks. In contrast, narrative studies focus solely on crisis episodes, adverse events, which have large and persistent effects on output. Our estimated effects of financial market disruptions are smaller than narrative studies like Romer and Romer (2017), however, consistent with the fact that some of the movements in financial distress identified by narrative studies are likely endogenous. We then use our model to revisit the effects of the financial crisis on output, in particular on the large gap that opened between output and its precrisis trend. To do so, we conduct a counterfactual model simulation based on parameters estimated with pre-2007 data, in which we turn off the financial shocks experienced over 2007–2008. We find that without the 2007–2008 financial shocks, the decline in output would have been a lot milder and only transitory. We conclude that a large fraction (over two-thirds) of the gap between current GDP and its precrisis trend was caused by the financial crisis. As additional evidence, we also consider the effects of financial shocks from U.K. data, and we obtain very similar conclusions. We follow the same identification strategy as in the United States, using data on the excess bond premium from Bleaney, Mizen, and Veleanu (2016) and the narrative measure of exogenous monetary policy changes from Cloyne and Hürtgen (2016). As with the United States, financial market disruptions have large and persistent effects on output, so that a large fraction of the gap between current U.K. output and its precrisis trend can be attributed to the financial crisis.